Between 1977 and 1984, estimates of hospital-acquired bloodstream infections caused by Candida species increased in the United States from 0.5 to 1.5 per 10,000 admissions (National Nosocomial Infection Study data). We examined crude and attributable mortality rates and excess length of stay in 88 closely matched pairs of cases and controls with illnesses occurring between July 1983 and December 1986. The crude mortality rates for cases and controls were 57% and 19%, respectively; thus the attributable mortality rate was 38% with a 95% confidence interval of 26% to 49%. The risk ratio was 2.94 with a 95% confidence interval of 1.95 to 4.43. The median length of stay was 48 days for all cases and 40 days for all controls. An analysis of the length of stay for the 34 matched pairs that survived showed a median of 70 days for cases and 40 days for controls. Candida bloodstream infections represented 10% of all nosocomial bloodstream infections in the period studied at our University Hospital; they are associated with a significant medical and economic burden well above that expected of the underlying diseases alone.